Sobre a questão da formalização do raciocínio abdutivo via sistemas dedutivos rotulados.

Abstract

The intention here is to develop a theory of machine learning (in particular, propositional abductive reasoning) using the framework of D. Gabbay's Labelled Deductive Systems (LDS), in the light of Gillies' perspective on the dichotomy 'deductive versus inductive logic' where controlled inference serves as the bridging notion. From this point of view, abductive reasoning, a topic that attracts much interest in A I and automated reasoning research, can be seen as a kind of controlled deduction where the control component will be represented by the labels of the LDS framework. This work investigates the possibility of treating both object and meta-level aspects of an abductive problem side by side: the set of possible explanations to an observed fact is generated deductively (the object level) and, after that , the most interesting explanation is chosen considering the preference criterion stated by the user (the meta-level). This way, i t is possible to give an abductive problem a more plausible solution. Moreover, since the preference criterion, which depends on the meaning of the labels, may vary, this characterization of abduction can easily be adapted to different problem domains.